DEEP-PNHG: Dynamic Entity-Enhanced Personalized News Headline Generation with Factual Consistency

This paper introduces DEEP-PNHG (Dynamic Entity-Enhanced Personalized Headline Generation), a novel framework addressing critical challenges in personalized news headline generation, namely dynamically evolving user interests, factual consistency, and informativeness. Existing methods often struggle with these aspects. DEEP-PNHG overcomes these limitations through an innovative architecture comprising a Dynamic User Interest Graph Module, an Entity-Aware News Encoder enriched with external knowledge, and a Fact-Consistent & Personalized Joint Decoder. This design enables the generation of headlines that are simultaneously tailored to dynamic user preferences, factually accurate, and highly informative. The framework’s key contributions include its ability to model evolving user interests, leverage external entities for robust factual understanding, and employ a joint decoding strategy enhanced by entity-level contrastive learning. Evaluated on a standard dataset, DEEP-PNHG consistently outperforms state-of-the-art baselines across personalization, factual consistency, and informativeness metrics. Our results demonstrate substantial improvements, positioning DEEP-PNHG as a significant advancement in generating engaging and trustworthy news headlines.